What it does
whisper-ctranslate2 is a command-line client for Whisper speech recognition that accepts the same options as OpenAI's Whisper command, so existing commands carry over. It runs on CTranslate2 and faster-whisper, which the README says is up to four times faster than openai/whisper at the same accuracy while using less memory, with batched inference adding more speed. Besides transcription and translation to English, it supports speaker identification through pyannote.audio (which needs a Hugging Face token), a voice activity detection filter, live transcription from a microphone, colour-coded confidence output, and loading your own fine-tuned Whisper models in CTranslate2 format. Quantization options such as int8 help on CPU. Install it with pip, or use the prebuilt Docker image that includes the small, medium and large-v2 models. It runs on x86-64 and ARM64 CPUs, and on NVIDIA GPUs when the CUDA 12 libraries are installed.
Capabilities
- Same command-line options as the original OpenAI Whisper client
- Faster inference through CTranslate2, with optional batching
- Speaker diarization through pyannote.audio
- Voice activity detection filter and live microphone transcription
- Loads your own fine-tuned models in CTranslate2 format
- Docker image with pre-loaded models
When to use it
- You already use the Whisper command line and want faster transcription
- You need speaker labels in JSON, VTT or SRT output
- You want to transcribe on a CPU or NVIDIA GPU, locally or in Docker
